A fingerprint including a plurality of ridges like a curved stripe pattern is characterized by permanence and uniqueness. Therefore, the fingerprint has been used for criminal investigation since long ago. In particular, matching using latent fingerprints left at a crime scene is effective investigating means. In recent years, many police agencies have introduced a fingerprint matching system using a computer.
However, many of images of latent fingerprints are low quality and include noise, and therefore judgment by a judge or automatization of the judgment is difficult. The images of the latent fingerprints include: an image of overlapped fingerprints in which ridges of two fingerprints overlap with each other; and an image including blur that forms into a curved stripe pattern. If one of the overlapped fingerprints is handled as a processing target, the other one can be regarded as background noise like a curved stripe pattern. The background noise like a curved stripe pattern is hereinafter referred to as curved stripe pattern noise. The blur forming a curved stripe pattern also corresponds to the curved stripe pattern noise.
The curved stripe pattern noise is the same as a fingerprint as a processing target (a target fingerprint) in that the both have curved stripe patterns. Accordingly, it is difficult to extract only a target fingerprint from the overlapped fingerprints, or remove blur forming a curved stripe pattern to prevent a target fingerprint from deteriorating.
An image processing method related to the present invention will be described below.
Non-patent literature 1 discloses a technique that applies a Fourier transformation to remove background noise. This technique is thought to be effective when periodic noise linearly appears in one direction, but have only a limited effect for curved stripe pattern noise. For example, in a region where a direction of a ridge of a target fingerprint and a direction of curved stripe pattern noise are close to each other, not only the curved stripe pattern noise but also the ridge of the target fingerprint may disappear. Further, even a ridge of a target fingerprint in a region where there is no noise may be deteriorated.
Patent literature 1 discloses a method for obtaining a direction distribution of a stripe pattern. In this method, an operator specifies a region and a direction indicating line in an image of the stripe pattern. The direction distribution of the stripe pattern in the region is obtained on the basis of the direction indicating line.
Also, various methods that extract directions and periodicity of fingerprint ridges, and perform a filtering process suitable for the directions and periodicity to enhance the fingerprint ridges are proposed. For example, Non-patent literature 2 and Patent literature 2 disclose such a method. However, such a method is thought not to be effective when directions and periodicity of ridges of a target fingerprint cannot be accurately extracted due to the influence of curved stripe pattern noise.
On the other hand, it is known that a local image enhancement method such as an Adaptive Contract Stretch method or Adaptive Histogram Equalization method is effective in removing local background noise. In the local image enhancement method, it is important to appropriately set a reference region for image enhancement.
Patent literature 3 discloses a technique that, by executing local image enhancement to a curved stripe pattern image including target fingerprint ridges and the curved stripe pattern noise based on directions of curved stripe pattern noise, a component of the curved stripe pattern noise is removed to enhance the target fingerprint ridges. In the local image enhancement, a reference region as a local region including a pixel of the curved stripe pattern image is set so as to be along a direction of the curved stripe pattern noise at a position of the pixel, and a density value of the pixel after the local image enhancement is calculated on the basis of a density histogram of the reference region.
According to the technique in Patent literature 3, when the number of curved stripe pattern noises is one (e.g., when the curved stripe pattern image includes, in addition to a target fingerprint, only one fingerprint that overlaps with the target fingerprint), the target fingerprint ridges can be enhanced. However, when the number of curved stripe pattern noises is two or more (e.g., when the curved stripe pattern image includes, in addition to the target fingerprint, two or more fingerprints that overlap with the target fingerprint), it is impossible to enhance only the target fingerprint ridges. Further, in a region where a direction of the curved stripe pattern noise is coincide with a direction of any of the target fingerprint ridges, the target fingerprint ridge is removed by the local image enhancement based on the direction of the curved stripe pattern noise.
Patent literature 4 discloses a stripe pattern direction determining system. The stripe pattern direction determining system uses a fact that a change in density along the same direction as stripes in a stripe pattern grayscale image is small, and a change in density along a direction orthogonal to the stripes is large, and thereby obtains a direction of a stripe in a pixel of the stripe pattern grayscale image.
Patent literature 5 discloses an image processing apparatus. The image processing apparatus divides an inputted fingerprint image into a plurality of small regions, performs a two-dimensional Fourier transformation for each of the small regions, and determines a direction of a fingerprint ridge in each of the small regions on the basis of a result of the two-dimensional Fourier transformation.